Test decision values.

This adds a test for decision values and fixes passing the pointer to
the decision value array.

src/LIBSVM.jl

@@ -399,9 +399,11 @@ function svmpredict(model::SVM{T}, X::AbstractMatrix{U}; nt::Integer = 0) where
 
     for i = 1:ninstances
         if model.probability
-            output = libsvm_predict_probability(cmod, nodeptrs[i], decvalues[:, i])
+            output = libsvm_predict_probability(cmod, nodeptrs[i],
+                                                Ref(decvalues, nlabels*(i-1)+1))
         else
-            output = libsvm_predict_values(cmod, nodeptrs[i], decvalues[:, i])
+            output = libsvm_predict_values(cmod, nodeptrs[i],
+                                           Ref(decvalues, nlabels*(i-1)+1))
         end
         if model.SVMtype == EpsilonSVR || model.SVMtype == NuSVR
             pred[i] = output

src/libcalls.jl

@@ -34,14 +34,14 @@ function libsvm_free_model(model::Ptr{SVMModel})
 end
 
 function libsvm_predict_probability(model::SVMModel, nodes::Ptr{SVMNode},
-        decisions::Vector{Float64})
+        decisions::Ref{Float64})
     return ccall((:svm_predict_probability, libsvm), Cdouble,
                  (Ref{SVMModel}, Ptr{SVMNode}, Ref{Float64}),
                  model, node, decisions)
 end
 
 function libsvm_predict_values(model::SVMModel, nodes::Ptr{SVMNode},
-        decisions::Vector{Float64})
+        decisions::Ref{Float64})
     return ccall((:svm_predict_values, libsvm), Float64,
                  (Ref{SVMModel}, Ptr{SVMNode}, Ref{Float64}),
                  model, nodes, decisions)

test/runtests.jl

@@ -48,7 +48,6 @@ end
     end
 end
 
-
 @testset "AbstractVector as labels" begin
     @info "test AbstractVector labels"
 
@@ -68,7 +67,6 @@ end
     @test ŷ == predict(model, Xtest')
 end
 
-
 @testset "JLD2 save/load" begin
     @info "JLD2 save/load"
 
@@ -156,5 +154,18 @@ end
     @test_throws ArgumentError svmtrain(rand(2, 5), ones(5); bad_params...)
 end
 
+@testset "Decision values" begin
+    X = [-2 -1 -1 1 1 2;
+         -1 -1 -2 1 2 1]
+    y = [1, 1, 1, 2, 2, 2]
+    d = [1.5 1.0 1.5 -1.0 -1.5 -1.5;
+         0.0 0.0 0.0  0.0  0.0  0.0]
+
+    model = svmtrain(X, y, kernel=Kernel.Linear)
+    ỹ, d̃ = svmpredict(model, X)
+
+    @test ỹ == y
+    @test d ≈ d̃
+end
 
 end  # @testset "LIBSVM"